AI Engineer · Azure Data Engineer (DP-203) · Python Developer
Building production multi-agent LLM systems, RAG pipelines, and cloud-scale ETL solutions
👆 Click to explore skills, projects, timeline, and career achievements as a live interactive page
- 🏆 Microsoft Certified Azure Data Engineer Associate (DP-203)
- 🤖 3+ years building production Generative AI — multi-agent platforms, RAG pipelines, LLM orchestration
- ☁️ 7+ years backend engineering — Azure Data Factory, ETL pipelines, cloud migration
- 🏛️ Delivered solutions for U.S. government clients — Contra Costa County, LA County, California State Agencies
- 🌐 Open to AI Engineer and Azure Data Engineer roles — Remote or Hyderabad
11 specialized AI agents analyse legacy codebases in parallel and generate full modernisation reports
- Agents: Architecture · Security · Code Generation · DevOps · Testing · Migration Planning · Documentation · UI/UX · Business Rules · Integration · Data Migration
- Backend: FastAPI + LangGraph + asyncio + SSE real-time streaming
- LLM Cascade: Claude → Gemini → Groq (LLaMA 3) → Ollama — automatic key rotation across 25+ API keys, zero-downtime fallback
- RAG: ChromaDB with AST file summaries, corpus deduplication, SHA-256 content fingerprinting
- Features: Spectator Mode, DOCX/MD export, self-correcting agents, batch context via Gemini's 1M-token window
| Project | What it does | Stack |
|---|---|---|
| 🤖 AppNova AI | 11 AI agents analyse legacy codebases in parallel | FastAPI · LangGraph · Claude · Gemini · ChromaDB |
| 🏛️ GovGenie | RAG-powered RFP generator — cuts bid writing by ~70% | LangChain · Ollama · ChromaDB · HuggingFace |
| 📄 ReferenceFiller | Resume → fills DOCX template in 2 min (was 45 min) | FastAPI · Ollama · ChromaDB · python-docx |
| 🎥 Video to Narrative | Surveillance footage → law-enforcement incident report | Flask · Whisper · ViT-GPT2 · Groq LLaMA 3 |
| 📊 Skill Matrix App | PDF/DOCX resume → HR skill matrix via RAG | Flask · Mistral · ChromaDB · Sentence Transformers |
| 🎬 Gemini YouTube Bot | Zero-input AI video creation + auto-publish | Gemini · MoviePy · gTTS · GitHub Actions |
📂 Project Details (click to expand)
Retrieves past proposal language from ChromaDB semantically, drafts government bid responses in company voice. Conversational chat interface with LangChain memory for follow-up clause questions. Reduced bid prep: 5+ days → under 24 hours
Upload resume → LLM extracts structured data → fills DOCX template via semantic field mapping. Session-based UUID stores, chunk-based text splitting, rotating log handlers. Replaced: 45-min manual task → 2-min automated workflow
OpenCV frame extraction → ViT-GPT2 captions → Whisper transcript → Groq LLaMA 3 synthesis. Returns structured JSON with timestamps, transcript, general summary, and formal law-enforcement narrative.
Fully local LLM (Mistral via Ollama) — no candidate data sent externally. RAG pipeline cuts hallucination to near zero vs baseline LLM-only approaches.
Gemini scripts → gTTS voiceover → MoviePy video → YouTube API auto-publish. Custom thumbnails, dynamic TextClip overlays. Runs entirely on GitHub Actions cron — no server needed.
30+ production ADF pipelines for U.S. government agencies
| Metric | Detail |
|---|---|
| Pipelines | 30+ ADF pipelines — person, address, offense, narrative, gang, property records |
| Performance | 50% faster data processing via optimised Mapping Data Flows |
| Migration | ARIES — 40+ tables, on-prem SQL Server → Azure Cloud, zero data loss |
| Automation | ADF validation + deduplication — 40% less manual review effort |
| Infrastructure | Azure VMs · Elastic Pools · Azure SQL · Data Lake Gen2 · Key Vault |
| Clients | Contra Costa County · LA County · California State Agencies |
